[Trainer] Fix optimizer step on PyTorch TPU

[cowanmeg]

What does this PR do?

Update optimizer step for TPUs to user self.optimizer.step() instead of xm.optimizer_step(self.optimizer).
AcceleratedOptimizer properly calls xm.optimizer_step on the optimizer (https://github.com/huggingface/accelerate/blob/main/src/accelerate/optimizer.py#L129).

This fixes a bug in transformers/trainer.py when using Pytorch on TPUs:
File "/usr/local/lib/python3.8/dist-packages/torch_xla/core/xla_model.py", line 471, in _fetch_gradients for param_group in optimizer.getstate()['param_groups']: KeyError: 'param_groups'

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@sgugger

[sgugger]

cc @muellerzr and @pacman100

[muellerzr]

Thanks! That is indeed the right solution

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[sgugger]

Thanks for checking Zach. Note to your future self: I think there is a lot of dupe code now ;-)

Thanks for your contribution @cowanmeg !

[pacman100]

Thank you @cowanmeg for the fix!